Skip to content

Dataset · nvidia

Detailed photo and video object descriptions from NVIDIA

nvidia/describe-anything-dataset

English multimodal dataset for training models to describe specific regions in images and videos in detail.

What's inside

The dataset was collected by researchers from NVIDIA, UC Berkeley, and UCSF to train the Describe Anything (DAM) model family. It contains 100,000 to 1,000,000 English examples covering images, video, and text annotations. Materials are packaged in tar archives and focus on localized descriptions of individual scene elements.

What you can do with it

  • Training VLM models (image-to-text and video-to-text) for detailed object description by selected region
  • Creating automated annotation and video content search systems
Tasks
Image → text, Video text → text
Languages
English
Size
100K–1M rows
Downloads this month
4.5K
Likes
60

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("nvidia/describe-anything-dataset")